DE102014209137A1 - Method and device for calibrating a camera system of a motor vehicle - Google Patents

Abstract

In a method and a device for calibrating a camera system of a motor vehicle, wherein the calibration parameters include the angles of rotation pitch angle, yaw angle and roll angle and the height of the camera above the roadway, the determination of the rotation angle from the determination of the vanishing point from a first optical flow between a first and second consecutive camera images and determining the height of the camera from a second optical flow between a first and a second consecutive camera image. In this case, a regular grid is placed over the first camera image to determine the first optical flow, in the second camera image correspondences of the regular grid are searched and the first optical flow is determined from the movement of the grid on the camera images.

Description

The invention relates to a method for calibrating a camera system of a motor vehicle according to the preamble of claim 1 and to a corresponding device according to the preamble of claim 11.

Today's motor vehicles with driver assistance systems are equipped with a large number of sensors, with many driver assistance systems relying on the data from cameras. To ensure reliable operation of the FAS, these cameras must be calibrated. For this purpose, according to the prior art, a complex static calibration performed in the factory. To save the static calibration, a robust online calibration system is needed, which after a short driving distance provides suitable calibration results such as the camera's three orientation angles and their height above ground.

In M. Siebeneicher: "An automatic calibration of any camera in vehicles based on optical flow and vanishing point determination", Master Thesis, Freie Universität Berlin 2008 , a method is explained which determines the orientation angles of the camera by evaluating the optical flux based on the motion of concise feature points and determining the vanishing point of the optical flow. To determine the vanishing point, the motion vectors of the feature points are intersected with each other, the intersections should theoretically coincide with the vanishing point, but this is only approximately the case due to measurement inaccuracies, so that the maximum of all intersections is taken as a vanishing point. In order to reduce measurement inaccuracies, the intersections are tracked with a particle filter over time and the imagewise resulting vanishing points are stabilized by RANSAC filtering.

To determine the roll angle from the horizon, the time course of the vanishing point is evaluated. Once a certain number of vanishing points have been determined in left and right turns, the horizon is determined by principal component analysis from a subset of all previously calculated vanishing points. By comparing the speed changes and the yaw angle, it is determined whether the vehicle is moving uniformly straight ahead and therefore the pitch and yaw angles can be determined above whether cornering is being performed. The height is determined separately in this procedure. However, the system is very computationally intensive, in particular by the feature determination and the elaborate filtering of the vanishing point.

This in J. Platonov et al: "Fully Automatic Camera-to-Vehicle Calibration", ATZ elektronik 2012, pp. 120-123 , presented method describes a technique for purely image-based determination of the camera orientation without a determination of the camera height. The method is based on the technique of visual motion estimation. A first method determines the vehicle motion between two adjacent images, and a second method considers the point correspondences between six images, but more specifically, the second method is significantly more computationally expensive. To determine the orientation angles, the camera movement is subdivided into the "straight-line motion" and "maneuver" classes. The pitch and yaw angle is estimated during the straight-line movement, the roll angle during the maneuver.

In E. Tapia et al: "A Note on Calibration of Video Cameras for Autonomous Vehicles with Optical Flow", Department of Mathematics and Computer Science, Series B Computer Science, Freie Universität Berlin, February 2013 , a method is presented, which determines the orientation of the camera only on the basis of camera images and with the help of the vehicle speed the camera height. In a manner not further described, the optical flux is determined and shown to intersect the flow vectors at vanishing point when traveling straight ahead. To determine the horizon, the ground plane is determined, with the aid of the vanishing point and the horizon, the three Euler angles can be determined. To determine the camera height, the vehicle moves forward, with the camera orientation being known and the camera looking at the road so that the roadway images in the camera image. The start and end points of a flow vector in the image are projected back into the world coordinate system. With this information, taking into account the vehicle speed and the driving time, the camera height can be calculated.

The publication EP 2 131 598 A2 relates to a stereo camera system and a method for determining at least one calibration error of a stereo camera system. With the aid of at least two individual cameras of the stereo camera system, an image sequence with images with images of a detection area in front of a vehicle during a journey of the vehicle along a roadway is respectively recorded. In this case, the images of the image sequences corresponding image data are generated. The generated image data are processed, wherein the course of at least one roadway edge of the roadway is determined. Starting from the determined course of at least one roadway edge of the roadway, at least one calibration error is determined.

The known systems are to calculate the position of all three angles of the camera orientation and the camera height above ground computationally intensive often with insufficient accuracy, especially in bad weather conditions and short driving distance with little cornering.

The invention is therefore based on the object of specifying a method and a device for calibrating a camera system of a motor vehicle, which allow a faster calibration of the camera system.

This object is achieved by a method having the features of claim 1 and by an apparatus having the features of claim 11. Preferred embodiments of the invention are subject of the dependent claims.

In the inventive method for calibrating a camera system of a motor vehicle, wherein the calibration parameters include the rotation angle pitch angle, yaw angle and roll angle and the height of the camera above the road, the determination of the rotation angle from the determination of the vanishing point of a first optical flux between a first and a second camera image and the determination of the height of the camera from a second optical flow between a first and a second consecutive camera image. In this case, a regular grid is placed over the first camera image to determine the first optical flow, in the second camera image correspondences of the regular grid are searched and the first optical flow is determined from the movement of the grid on the camera images.

Thus, the optical flux between each two images is determined by a regular grid is placed over the first image and the raster points of this grid are searched for example by means of the Lucas-Kanade method in the second image.

Preferably, the distances of the screen dots can be parameterized. In this way, scalability of the optical flow calculation can be achieved. The smaller the distance between the halftone dots, the longer the calculation takes, but this gives a more accurate vanishing point estimate.

On the basis of the determined first optical flow, the epipoles are calculated in both camera images and the epipoles form the basis of the vanishing point estimate. The determination of the epipoles can be done by calculating the optical flow based fundamental matrix.

More preferably, the vanishing point is determined for the pitch angle and yaw angle determination from the two epipoles of the two camera images, if the distance of the epipole is within a first predetermined interval. The determination of the pitch angle and the yaw angle takes place during a straight ahead, so that the two epipoles are close to each other. From the two epipoles, for example, the mean is formed, which then enters as the current vanishing point in the estimation of pitch and yaw angle.

The determined vanishing points are preferably stored over time and a time-stable vanishing point for determining the pitch angle and yaw angle is determined from the set of vanishing points determined. In this case, a storage interval can be specified, which moves with the current time. In this way, old vanishing points can be discarded. Furthermore, the number of determined vanishing points that are necessary for estimating the time-stable vanishing point can be determined via the storage interval.

More preferably, the determined epipoles of two camera images are used for roll angle determination when the distance of the epipole is within a second predetermined interval, wherein a roll angle is estimated from the position of the epipoles in the respective camera image. In other words, if the two epipoles are "far apart", the vehicle is cornering and a current roll angle can be determined. In order to distinguish for which angle of rotation the current epipoles are used, the two predetermined intervals are disjoint.

Preferably, the determined roll angles are collected over time and divided into predetermined groups as a function of the yaw rate of the motor vehicle. In this case, preferably the expiry time and the number and definition of the groups can be parameterized.

More preferably, an average group roll angle is formed for each predetermined group of stored roll angles when the number of measurements in each group has reached a minimum number. For example, the final roll angle used for calibration may be formed as an average of the mean group roll angles over the groups.

Preferably, in order to determine the second optical flow in a section of a first image of the roadway in front of the motor vehicle, a rectangle is placed in plan view whose position is determined in the second image, wherein the second optical flow is formed by the movement of the rectangle in the two images the rectangle movement is calculated the height of the camera. In this way, a greatly simplified second optical flow is determined, wherein the finding of the rectangle in the second image can be done by block matching, so that the movement of the rectangle and thus the second optical flow can be determined. The rectangle can be parameterized with the parameters height and width of the rectangle.

More preferably, a temporal histogram of the detected heights of the camera is created for a plurality of first and second camera images is derived with the achievement of a minimum number of measurements from the histogram of an estimated height. Using a histogram improves the elevation estimation, as the detected motion of the rectangle is heavily noise-locked.

• – eine Einrichtung zur Bestimmung eines ersten optischen Flusses zwischen einem ersten und einem zweiten aufeinanderfolgenden Kamerabild,
• – eine Einrichtung zur Bestimmung von Fluchtpunkten aus dem ersten optischen Fluss,
• – eine Einrichtung zur Bestimmung der Drehwinkel aus ermittelten Fluchtpunkten,
• – eine Einrichtung zur Bestimmung eines zweiten optischen Flusses zwischen einem ersten und einem zweiten aufeinanderfolgenden Kamerabild, und
• – eine Einrichtung zur Bestimmung der Höhe der Kamera aus einem zweiten optischen Fluss, wobei zur Bestimmung des ersten optischen Flusses ein regelmäßiges Raster über das erste Kamerabild gelegt wird, im zweiten Kamerabild Korrespondenzen des regelmäßigen Rasters gesucht werden und der erste optische Fluss aus der der Bewegung des Raster über den Kamerabilder bestimmt wird.

The device according to the invention for calibrating a camera system of a motor vehicle, which is set up and designed to carry out the method explained above, comprises:

• A device for determining a first optical flow between a first and a second consecutive camera image,
• A device for determining vanishing points from the first optical flow,
• A device for determining the angles of rotation from determined vanishing points,
• A device for determining a second optical flow between a first and a second consecutive camera image, and
• - A device for determining the height of the camera from a second optical flux, wherein for determining the first optical flow, a regular grid is placed over the first camera image, in the second camera image correspondences of the regular grid are searched and the first optical flow from the movement of the grid over the camera images is determined.

More preferably, the device for determining the vanishing point on the basis of the determined first optical flow calculates the epipoles in both camera images and uses the epipoles for vanishing point estimation.

Preferably, in order to determine the second optical flow in a section of a first image of the road in front of the motor vehicle, a rectangle is placed in plan view, which is determined in the second image, wherein the second optical flow is formed by the movement of the rectangle in the two images the rectangle movement is calculated the height of the camera.

More preferably, the means for determining the camera height creates a temporal histogram of the measured heights of the camera for a plurality of first and second camera images, and derives an estimated height upon reaching a minimum number of measurements from the histogram.

Advantageously, the method according to the invention or the corresponding device is able to determine the yawing and pitching angles after a short driving distance when driving straight ahead, as well as after a short journey with at least one left and one right turn. Furthermore, the camera height above ground is determined in parallel, so that a calibration of the camera takes place at a significantly shorter convergence time. There are no dependencies on predefined markings, structures or shapes, in other words, the described method or device works in any environment. The environment must have only a minimum of structure that can be detected by the camera, and allow at least one steering maneuver of the vehicle.

In addition, no elaborate feature extraction is performed and the explained approach is constructed so that it can be implemented suitable for ECUs.

A preferred embodiment of the invention will be explained below with reference to the drawings. It shows

1 the coordinate systems for the definition of extrinsic camera parameters,

2 the structure of the calibration system in a schematic representation, and

3 a schematic representation of the height estimation.

1 shows the world coordinate system (O, X, Y, Z) with the zero point O and the spatial axes X, Y and Z, in which the motor vehicle moves, and the coordinate system (C, X _C , Y _C , Z _C ) of the camera K , in the 1 is shown schematically, with the zero point C and the axes X _C , Y _C and Z _C. In this case, the camera coordinate system (C, X _C , Y _C , Z _C ) relative to the world coordinate system (O, X, Y, Z) of the vehicle can be moved and rotated about the three axes X _C , Y _C and Z _C. The possible rotation angles are in 1 shown as pitch angle θ, yaw angle φ and roll angle Ψ. For calibration of the camera of a motor vehicle, as already mentioned, the parameters height of the camera above the roadway, pitch angle θ, yaw angle φ and roll angle Ψ must be determined.

The world coordinate system (O, X, Y, Z) and the camera coordinate system (C, X _C , Y _C , Z _C ) may differ in position and orientation from each other. The difference in orientation is indicated by a rotation about the respective axis, the camera K rotating about three axes X _C , Y _C , Z _C and describing each rotation through a corresponding rotation matrix R _Nick , R _Gier and R _Roll in the usual way can be. The complete rotation matrix R is a series of individual rotation matrices and is formed as follows: R = R _Roll R _Nick R _Gie and corresponds to a rotation about the Y-axis, followed by a rotation about the X-axis and finally about the Z-axis.

Together with the translation vector C = (C _x , C _y , C _z ) ^T , which describes the zero point of the camera coordinate system in the world coordinate system and indicates the difference of the positions of the two coordinate systems, the uniform transformation matrix can be formed, which converts the world coordinate system into the camera coordinate system ,

By means of the vanishing point, yaw angle θ and pitch angle φ can be determined from the complete rotation matrix R during a rectilinear movement, since the roll angle Ψ is zero for a rectilinear movement. For a determination of the roll angle ein a vanishing point is not sufficient, but the roll angle Ψ from the course of the vanishing point can be determined during cornering, as the vanishing point changes its position proportional to the roll angle Ψ. Details of the theoretical fundamentals can be found, for example, in the aforementioned master thesis by M. Siebeneicher.

A determination of the orientation or rotation angle therefore requires the most accurate determination of the vanishing points used and their movement, the determination of which in turn depends on an optimal determination of the optical flow.

2 is a schematic representation of the calibration system KP for determining the above-mentioned calibration parameters of the camera of a motor vehicle, wherein at the beginning a brief overview of the components of the calibration KP takes place and subsequently the individual components are explained in detail.

Input variables of the calibration system KP are the recorded camera images B1 and B2 as well as the ego data ED of the vehicle. Here, under ego data of the vehicle, general vehicle data such as, for example, tachometer values, steering angles, odometry data and position data, as well as acceleration data, will be the yaw angle and pitch rates. The list is not to be considered as complete. In this case, the ego data can be supportive for the calibration system in order to identify, for example, driving conditions that are particularly favorable for the calibration process.

The calibration of the vehicle camera with the aid of the calibration system KP is based on the determination of the optical flow within the images of the camera. For this purpose, motion vectors are determined via the image with the aid of features of the camera images B1, B2 in the first optical flow module O1 of the calibration system KP, the motion vectors being determined based on a grid arranged uniformly over the first image.

Based on the motion vectors determined in the first optical flux module O1, an estimate of the vanishing point based on geometrical properties takes place in the vanishing point module FP. The Vanishing point estimation of the vanishing point module FP is used in a subsequent angle determination module WB for determining the three angles of rotation, wherein the angle determination module WB has the interconnected modules yaw angle determination GW, roll angle determination RW and pitch angle determination NW.

Separately, in a second optical flux module, the determination of motion vectors of a greatly simplified optical flow in a plan view of an image section of the road ahead of the vehicle takes place. The height of the camera above the roadway in the height estimation module HS can be estimated from the determined motion vectors of the roadway characteristics as well as ego data and the determined pitch angle θ.

The following explains the details of the individual components:

FIRST OPTICAL RIVER:

The determination of the first optical flux O1 takes place in each case between two images B1, B2 of the camera. In this case, no dense flux field is determined, but a regular grid is placed over the first image B1. By means of, for example, the known Lucas-Kanade method, correspondences to the raster in the second image B2 are searched for. The distances of the grid points can be parameterized, whereby the calculation of the optical flow is very scalable. In other words, the smaller the pitch of the halftone dots, the greater the running time of the calculation, but the accuracy of the vanishing-point estimation based thereon increases.

A further advantage of the grid approach over a feature-based flow determination is that conspicuous features are mostly found on the contours of moving objects, and these strangers moved objects then distort the vanishing point.

Vanishing Point ESTIMATE:

Based on the optical flow, the fundamental matrix of the epipolar geometry is calculated. This provides the respective epipole in both images B1, B2 considered. When traveling straight ahead, the two epipoles must be very close to each other, whereby the distance can be parameterized. It is the mean of the two epipoles formed that forms the current vanishing point and is passed to the pitch and yaw angle estimation.

The following conditions apply to the distance "very close to each other" of the two epipoles for the pitch and yaw angle determination:
minimum X-distance of the pixels relative to the image width = 0,
maximum X-distance of the pixels relative to the image width = approx. 0.02,
minimum Y-distance of the pixels relative to the image height = 0,
maximum Y-distance of the pixels relative to the image height = approx. 0.01.

If the distance of the epipoles is greater than the abovementioned maximum distances, the determination of the mean values of pitch and yaw angles is much more difficult to impossible, since more and stronger outliers must be taken into account.

If the vehicle is cornering, then the two epipoles of the images B1, B2 must be further apart. If this condition is fulfilled, then the two epipoles are passed on to the roll angle estimation RW.

For roll angle determination, the distances of the epipoles must meet the following conditions:
minimum X-distance of the pixels relative to the image width = approx. 0.02
maximum X-distance of the pixels relative to the image width = approx. 0.1
minimum Y-distance of pixels relative to image height = 0.01
maximum Y-distance of the pixels relative to the image height = approx. 0.05

This procedure can be carried out in comparison to known methods with much less running time and in fewer steps.

ANGLE DETERMINATION:

DETERMINATION OF NICKWINKEL AND YELLOW ANGLE.

The vanishing points determined in the vanishing point estimate are collected over time, discarding old vanishing points. The number and expiry time can be parameterized. If there are enough vanishing points, a time-stable vanishing point is estimated from the set of highly dynamic vanishing points using, for example, the well-known RANSAC algorithm. This time-stable vanishing point is used to determine the pitch angle and the yaw angle Ψ in the corresponding modules NW, GW.

ROLL ANGLE ESTIMATE

• – hohe Gierrate nach links,
• – mittlere Gierrate nach links,
• – hohe Gierrate nach rechts, und
• – mittlere Gierrate nach rechts.

The epipoles collected during cornering are used directly to determine a possible roll angle via their position in the image. These highly dynamic roll angles are collected over time and divided into groups, whereby the decay time, number and groups can be parameterized. Preferably, four groups are used, namely:

• High yaw rate to the left,
• - mean yaw rate to the left,
• - high yaw rate to the right, and
• - mean yaw rate to the right.

The ranges of the yaw rate α given below for the definition of the four groups mentioned above have proven to be successful: Range for slow cornering left: 0 ° / s <= α <8 ° / s (mean yaw rate left) Area for fast cornering left: 8 ° / s <= α <45 ° / s (high yaw rate left) Slow cornering area on the right 0 ° / s> = α> -8 ° / s (average yaw rate right) Right turn corner area: -8 ° / s> = α> -45 ° / s (high yaw rate on the right)

There must be a minimum number of measurements in each group. If this is the case, the mean value is calculated for each group and then the average is again formed over these four mean values. This average is used to determine the roll angle RW.

HEIGHT ESTIMATE

In the standard implementation, a very simplified optical flow is determined. In the first step, a plan view is calculated from an image section directly in front of the vehicle, wherein the image section can be parameterized. Within this section, a rectangle is placed in the first image, which must be found again by means of block matching in the second image, taking into account the parameters height and width of the rectangle. The movement of the rectangle in the picture forms the optical flow.

For the rectangle designated as a template box, the following boundary conditions apply in the standard implementation:
Template box width relative to TopView width = approx. 0.9
Template box width relative to template box width = approximately 0.4

The template box is positioned in the TopView image of the first frame in such a way that the upper edge of the template box is in the middle of the upper edge of the image. H. left and right are about 5% edge compared to the top view width.

HEIGHT DETERMINATION

The movement of the rectangle in the image determined in the first step is converted directly into a real movement of the vehicle, whereby the value is strongly underlaid by noise. Therefore, a histogram is constructed over time with the lowest and highest values of the histogram as well as the number of bins. Again, the measurements can expire with their age. If there is a minimum number of measurements in the histogram, then the N adjacent bins containing most of the measurements are searched. Subsequently, the median is formed over these bins. The result is then the estimated height.

kann durch die Anwendung einfacher Trigonometrie aus den bekannten Größen die Höhe h zu

bestimmt werden. The basic procedure for determining the height is based on the 3 explained, wherein the 3 a simple hole camera model is the basis and the height determination is discussed here based on the movement of the middle pixel. If the vehicle (not shown) moves rectilinearly at a known current speed v and the camera's pitch angle is known, the camera height h relative to the roadway F can be determined with the aid of the optical flow. The optical flow provides the velocity v _P at which the pixel moves in the center of the camera. 3 1 shows the geometry for determining the camera height h and represents two different times t and t + 1, for which the projected image center in the camera deflection plane k (t) and K (t + 1) shown schematically at times t and t + 1 is considered with the camera having the focal length f. The values for the distance traveled v, the pitch angle θ of the camera, the focal length f of the camera and the velocity v _{P of} the center pixel in the image are known. From the obvious relationship

By using simple trigonometry, the height h can be obtained from the known quantities

be determined.

LIST OF REFERENCE NUMBERS

• O

  Zero point world system

  X

  X axis

  Y

  Y-axis

  Z

  Z-axis

  C

  Zero point camera system = translation vector

  C _X

  X component translation vector

  C _Y

  Y component translation vector

  C _z

  Z component translation vector

  X _C

  X-axis camera

  Y _C

  Y-axis camera

  Z _C

  Z-axis camera

  θ

  pitch angle

  φ

  yaw

  Ψ

  roll angle

  KP

  Determination of calibration parameters

  ED

  Ego data

  B2

  Camera picture 1

  B1

  Camera picture 2

  O1

  Optical flow 1

  O2

  Optical flow 2

  FP

  Vanishing Point Estimate

  WB

  angle determination

  GW

  Yaw angle determination

  RW

  Roll angle determination

  northwest

  Nick angle determination

  hs

  height estimate

  F

  roadway

  H

  Height of the camera above the roadway

  v

  vehicle speed

  v _p

  pixel speed

  f

  focal length

  K (t)

  Image plane of the camera at time t

  K (t + 1)

  Image plane of the camera at time t + 1

  v '

  auxiliary variable

  f '

  auxiliary variable

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• EP 2131598 A2 [0007]

Cited non-patent literature

• M. Siebeneicher: "An automatic calibration of any camera in vehicles based on optical flow and vanishing point determination", Master Thesis, Freie Universität Berlin 2008 [0003]
• J. Platonov et al: "Fully Automatic Camera-to-Vehicle Calibration", ATZ elektronik 2012, pp. 120-123 [0005]
• E. Tapia et al: "A Note on Calibration of Video Cameras for Autonomous Vehicles with Optical Flow", Department of Mathematics and Computer Science, Series B Computer Science, Freie Universität Berlin, February 2013 [0006]

Claims (14)

A method for calibrating a camera system of a motor vehicle, wherein the calibration parameters include the rotation angle pitch angle, yaw angle and roll angle and the height of the camera over the road, wherein the determination of the rotation angle from the determination of the vanishing point of a first optical flow between a first and a second camera image deriving the determination of the height of the camera from a second optical flow between a first and a second camera image is guided, characterized in that for determining the first optical flow, a regular grid is placed over the first camera image, in the second camera image correspondences of the regular Rasters are searched and the first optical flux from the movement of the grid is determined on the camera images. A method according to claim 1, characterized in that the distances of the screen dots are parameterized. Method according to one of the preceding claims, characterized in that on the basis of the determined first optical flow, the epipoles are calculated in both camera images and the epipoles are used for vanishing point estimation. A method according to claim 3, characterized in that the vanishing angle and yaw angle determination from the two epipholes of the two camera images of the vanishing point is determined when the distance of the epipole is within a first predetermined interval. A method according to claim 4, characterized in that the determined vanishing points are collected for a predetermined time interval and from the set of vanishing vanishing points a time-stable vanishing point for determining pitch angle and yaw angle is determined. A method according to claim 3, characterized in that the two epipoles of two camera images are used for roll angle determination when the distance of the epipole is within a second predetermined interval, wherein from the position of the epipoles in the respective camera image, a roll angle is estimated. A method according to claim 6, characterized in that the determined roll angles are collected over time and divided into predetermined groups as a function of the yaw rate of the motor vehicle. A method according to claim 7, characterized in that , for each group of stored roll angles, an average group roll angle is formed when the number of measurements in each group reaches a minimum number, the final roll angle being formed from the mean group roll angles of the groups. Method according to one of the preceding claims, characterized in that for determining the second optical flow in a section of a first image of the road in front of the motor vehicle top view a rectangle is placed, which is determined in the second image, wherein the second optical flow through the movement of the Rectangles is formed in the two images and from the rectangle movement the height of the camera is calculated. A method according to claim 9, characterized in that a temporal histogram of the measured heights of the camera for a plurality of first and second camera images is created and with the achievement of a minimum number of measurements from the histogram, an altitude estimate is derived. Device for calibrating a camera system of a motor vehicle, which is set up and designed to carry out the method according to one of the preceding claims, having a device for determining a first optical flow between a first and a second consecutive camera images, a device for determining vanishing points from the first optical flux, means for determining the angles of rotation from the determined vanishing points, means for determining a second optical flow between a first and a second consecutive camera images, and means for determining the height of the camera from a second optical flow, characterized in that for determining the first optical flow, a regular grid is placed over the first camera image, in the second camera image correspondences of the regular grid are searched and the first optical flow from the movement of the Ras is determined over the camera images. Apparatus according to claim 11, characterized in that the means for determining the vanishing point on the basis of the determined first optical flow calculates the epipoles in both camera images and uses the epipoles for vanishing point estimation. Apparatus according to claim 11 or 12, characterized in that for determining the second optical flow in a section of a first image of the roadway in front of the motor vehicle in plan a rectangle is placed, which is determined in the second image, wherein the second optical flow through the movement of the rectangle formed in the two images and from the movement of the rectangle the height of the camera is calculated. Apparatus according to claim 13, characterized in that the means for determining the camera height creates a temporal histogram of the determined heights of the camera for a plurality of first and second camera images and derives a height estimate with the achievement of a minimum number of measurements from the histogram.

Images